Counterfactuals and causability in explainable artificial intelligence: Theory, algorithms, and applications

Deep learning models have achieved high performance across different domains, such as medical decision-making, autonomous vehicles, decision support systems, among many others. However, despite this success, the inner mechanisms of these are opaque because their internal representations too complex for a human to understand. This opacity makes it hard understand how or why predictions deep models. There has been growing interest in model-agnostic methods that make more transparent and explainable humans. Some researchers recently argued machine achieve human-level explainability, needs provide causally understandable explanations, also known causability . A specific class algorithms potential counterfactuals paper presents an in-depth systematic review diverse existing literature on artificial intelligence (AI). We performed Latent Dirichlet topic modelling analysis (LDA) under Preferred Reporting Items Systematic Reviews Meta-Analyses (PRISMA) framework find most relevant articles. yielded novel taxonomy considers grounding theories surveyed algorithms, together with underlying properties applications real-world data. Our research suggests current counterfactual AI not grounded causal theoretical formalism and, consequently, cannot promote decision-maker. Furthermore, our findings suggest explanations derived from popular spurious correlations rather than cause/effects relationships, leading sub-optimal, erroneous, even biased explanations. Thus, advances new directions challenges promoting approaches AI. • survey XAI. set systems Opportunities based modelagnostic approaches.